Waldek Hebisch wrote:
> Look at the following program:
>
> program pack2;
> type pr = packed record r: record j : integer end end;
> var v : pr;
> procedure foo(var i : integer);
> begin
> end;
> begin
> foo(v.r.j);
> end
> .
>
> I think this program _should_ be illegal (otherwise possibilities
> to pack pr would be quite limited), but I can not find full
justification
> based on ISO 7185 or ISO 10206. Namely, 6.6.3.3 (in ISO 7185) and
> 6.7.3.3 (in ISO 10206) tells us that components of packed records
> should not be passed as var parameters. But "j" is a component of "r",
> which is not a packed record. The question is if "j" is a
> component of "v"? I was not able to find any definition of
> component in the standards, but it seems that in the standards
> component means what could be otherwise called "direct component".
> If "j" is not a component of "v", then AFAICS the program above
> is legal...
>
> Any comments?
>
I compiled it as:
PC compiler shell vs. 1.12.00 Copyright (C) 2005 S. A. Moore
Reading instruction file c:\ip\windows\i80386\bin\pc.ins
Building C:\TEST\test
Pascal parser vs. 1.12.01 Copyright (C) 2005 S. A. Moore
Processing: C:\TEST\test.pas
*** C:\TEST\test.pas [6:5] Symbol 'i' defined but never referenced ***
foo(v.r.j);
^
*** C:\TEST\test.pas [8:9] Packed component cannot appear as variable
parameter
***
Errors this compilation: 2
No output file was generated
Function complete
Build has errors, terminating
Sooo, basically IP considers any element of a packed, however deeply
nested to be subject to that restriction. I think that rule is
intuitive.
I suspect the 6.6.3.3 limitation key is if by "components" of a record
that also includes components of components of components.. etc.
I'll see if I can't find a concrete quotation in 7185 for ya tomorrow.
|
